Abstract：Most query recommendations based on the search log are with the aspect of popularity.These kinds of query recommendations methods ignore the difference in search demand caused by different backgrounds of users.Sometimes，the query recommendations method based on popularity could not provide high quality service even the users have a lot of search record in the log.So，a balanced solution is proposed in this paper，serving as a search recommendation strategy which is popularity versus similarity providing recommendation service near to personalized.Intensive behavior block based on query-flow graph are discovered from search records based on community structure detection to construct representative user behavior model which describes the search background of users and introduce modularity which measures the strength of user behavior.The popularity，the degree of support，the degree of membership and the strength of user behavior are proposed to produce the result of query recommendations.Extensively experiments are conducted on a real query log，and the results show that this method can reduce the risk of recommendation failure and improve the satisfaction of users for recommendation.
2016, Vol. 37 
